Uncertainty analysis of human EEG spectra: a multivariate information theoretical method for the analysis of brain activity

Short time Fourier power spectra yield activity related data of the EEG. The mean power densities of frequency bands have proved a reliable descriptive parameter of that activity. Due to their distributional properties of these data their analysis by conventional statistics is restricted to nonparametric univariate or at most bivariate tests. In order to account for the many-relational characteristics of EEG data, uncertainty analysis was employed and specific data transformations were developed. This technique was applied to some of our experimental records. The results confirm prior knowledge of the functional diversity of brain areas and demonstrate the general applicability of uncertainty analysis to this problem. In addition, the spatio-temporal coordination of these activities suggests a systems interpretation: These neuronal processes follow the self-organizing principle.